Abstract: In order to reduce the identification error of rail vehicle vertical wheel-rail force in time domain, time-domain methods based on machine learning correction are carried out for vertical wheel-rail force identification. Firstly, the vehicle dynamics simulation model is established, and the axle box accelerations and vertical wheel-rail forces are obtained under the speed of 250 km/h with random track irregularity as the excitation. Secondly, the dynamic load identification models corresponding to Green function method and state space method are established. The initial value error of state space method is analyzed, and the polynomial fitting method is introduced to correct the trend term error. The performances of the two methods in calculation accuracy and efficiency are compared. Then, aiming at the identification error existing in the time domain methods, it is proposed to use NARX (nonlinear autoregressive models with exogenous inputs) model to train and predict the identification error, so as to reduce the influence of factors such as incomplete response observation and observation noise in the model, and then correct the identification results of the time-domain methods to improve the identification accuracy. Finally, the correctness of the methods are verified by a 10-degree-of-freedom rail vehicle vertical dynamics model. The results show that the two methods have high identification accuracy for vertical wheel-rail forces of rail vehicles, and for each wheelset, each method has its own advantages and disadvantages; in terms of computational efficiency, state space method is better than Green function method; the two time-domain methods corrected by NARX model are effective in identifying the vertical wheel-rail forces of rail vehicles, and the Pearson correlation coefficients between the simulation and identification values are greater than 0.99, belonging to extremely strong correlation. The machine learning error correction method based on NARX model can effectively improve the time domain identification accuracy, and can provide reference for the subsequent prediction of wheel-rail force of rail vehicles, with strong engineering application value.